PROCESS MODELS DISCOVERY AND TRACES CLASSIFICATION: A FUZZY-BPMN MINING APPROACH.

Transcription

1 Journal of International Technology and Information Management Volume 26 Issue 4 Article PROCESS MODELS DISCOVERY AND TRACES CLASSIFICATION: A FUZZY-BPMN MINING APPROACH. Kingsley Okoye Dr University of East London, k.okoye@uel.ac.uk Usman Naeem Dr University of East London, UK, u.naeem@uel.ac.uk Syed Islam Dr University of East London, UK, syed.islam@uel.ac.uk Abdel-Rahman H. Tawil Dr University of East London, UK, A.R.Tawil@uel.ac.uk Elyes Lamine Dr Université de Toulouse, Mines-Albi, France, Elyes.Lamine@mines-albi.fr Follow this and additional works at: Part of the Artificial Intelligence and Robotics Commons, Business Intelligence Commons, Computer and Systems Architecture Commons, Data Storage Systems Commons, Management Information Systems Commons, Numerical Analysis and Scientific Computing Commons, Software Engineering Commons, Technology and Innovation Commons, and the Theory and Algorithms Commons Recommended Citation Okoye, Kingsley Dr; Naeem, Usman Dr; Islam, Syed Dr; Tawil, Abdel-Rahman H. Dr; and Lamine, Elyes Dr (2017) "PROCESS MODELS DISCOVERY AND TRACES CLASSIFICATION: A FUZZY-BPMN MINING APPROACH.," Journal of International Technology and Information Management: Vol. 26 : Iss. 4, Article 1. Available at: This Article is brought to you for free and open access by CSUSB ScholarWorks. It has been accepted for inclusion in Journal of International Technology and Information Management by an authorized editor of CSUSB ScholarWorks. For more information, please contact scholarworks@csusb.edu.

2 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach. Kingsley Okoye Usman Naeem Syed Islam Abdel-Rahman H. Tawil School of Architecture Computing & Engineering, University of East London, Docklands Campus, E16 2RD {K.Okoye, U.Naeem, S.Islam, A.R.Tawil}@uel.ac.uk UNITED KINGDOM Elyes Lamine Université de Toulouse, Mines-Albi, CGI, Campus Jarlard, Albi Cedex 09, France Elyes.Lamine@mines-albi.fr FRANCE ABSTRACT The discovery of useful or worthwhile process models must be performed with due regards to the transformation that needs to be achieved. The blend of the data representations (i.e data mining) and process modelling methods, often allied to the field of Process Mining (PM), has proven to be effective in the process analysis of the event logs readily available in many organisations information systems. Moreover, the Process Discovery has been lately seen as the most important and most visible intellectual challenge related to the process mining. The method involves automatic construction of process models from event logs about any domain process, and describes causal dependencies between the various activities as performed within the process execution environment. In principle, one can use process discovery to obtain process models that describes reality. To this end, the work in this artcle presents a Fuzzy-BPMN mining approach that uses training events log representing 10 different real-time business process executions to provide a method for discovery of useful process models, and then cross-validating the derived models with a set of test event logs in order to measure the accuracy and performance of the employed approach. The method focuses on carrying out a classification task to determine the traces, i.e. individual cases that makes up the test event logs in order to determine which traces that can be replayed by the original model. Thus, the paper aim is to provide a technique for process models discovery which is as good in balancing between overfitting and underfitting International Information Management Association, Inc ISSN: On-line Copy

3 Journal of International Technology and Information Management Volume 26, Number as it is able to correctly classify the traces that can be replayed (i.e allowed) or non-replayable (disallowed) by the model. In other words, the study shows through the Fuzzy-BPMN replaying notation and the series of validation experiments - how given any classified trace (for the test events log) and discovered process model (the training log) it can be unambiguously determined whether or not the traces found can be replayed on the discovered model. KEYWORDS: process mining, process discovery, classifiers, fuzzy models, BPMN notation, event logs, classification, process models INTRODUCTION The need for novel approaches in design and integration of computational intelligence and technologies into everyday (e.g. business) processes, have sprout new insights and unceasing research investigations particularly on how to exploit such tools for use in improving the various organisational processes (Van der Aalst, 2016; Van der Aalst et al, 2010). In recent years, a common challenge with many of the business processes has been on how to develop intelligent systems and/or techniques that can provide platforms for exploring the additional, and most often, the monotonous tasks of managing the entire operational process and quality of information - by providing understandable and useful insights on the best possible ways to make the envisioned information explicable in reality using the underlying events log recorded in the IT systems. Most organizations have invested in projects to model their various operational processes. However, majority of the derived process models are often unfitting, non-operational, or represents a form of reality that are pointed towards comprehensibility rather than covering the entire actual business process complexities. Therefore, the ability to mine useful or worthwhile knowledge from the readily extracted datasets in current information systems appears to be a challenge, due to the exponential increase in the volume of data that is generated. In consequence, this has spanned the need for a richer and advanced description of real-time processes that allows for flexible exploration of the large volumes of data targeted at improving the system performance and analysis. Even more, researchers (Dou, et al., 2015; de Medeiros & Van der Aalst, 2009; Van der Aalst, 2016) have shown that a better way of attaining a closer look at any organisation s operational process is to consider the events log that are readily available in its process-base or IT systems. Perhaps, an accurate exploration and/or analysis of the events log could provide vital and valuable information with regards to the quality of support being offered for the so-called organizations and their information systems at large. For example, revealing the underlying relationships International Information Management Association, Inc ISSN: On-line Copy

4 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al the process elements or individual actors share amongst themselves within the knowledge-base. Such process-related analysis, often allied to the process mining, means there is also need for tools and techniques that are capable of extracting valuable information from the event logs about the real-time processes. Practically, there are two main drivers for such growing interest in the process mining field. On the one hand, more and more events are being recorded, thus, providing detailed information about history of processes as they happen in reality. On the other hand, there is need to improve and support business processes in a competitive and rapidly changing environment (Van der Aalst, 2016). Thus, process mining (PM) means extracting valuable, process-related information from event logs about any real time process. Recently, the Process Mining (Van der Aalst, 2016) has become a valuable technique used to discover such meaningful information from the event data logs. Besides, the PM field combines techniques from computational intelligence which has been lately considered to encompass artificial intelligence (AI) or even the latter, augmented intelligence (AIs) systems, and data mining (DM) to process modelling, as well as several other disciplines to analyze the events logs. Indeed, since the PM techniques builds on computational intelligence and data mining techniques, which has led to its significant influence on how process owners and analysts perceive and analyse the readily available large volumes of data captured from their various IT systems. Besides, a greater number of the resulting models and methods tends to support not just machine-readable systems but also machineunderstandable systems. By machine-understandable systems we refer to methods that are developed not just for representing information in formats that can be easily understood by humans, but also for creating applications and/or systems that trails to inclusively process the information that they contain or supports. Furthermore, the Classification - according to (Han and Kamber, 2005) is one of the most universally data mining technique that aims at finding models or functions that describes or distinguishes data attributes or concepts. Specifically, the authors in (Elhebir and Abraham, 2015) notes that pattern discovery algorithms makes use of statistical and machine-learning techniques to build models that predicts behaviour of captured datasets, and concedes that one of the most pattern discovery techniques used to extract knowledge from pre-processed data is classification. The authors observe that most of the existing classification algorithms attains good performance for specific problems but are not robust enough for all kinds of discovery problems, and further propose that combination of multiple classifiers (i.e. Hybrid Intelligent Systems (HIS) such as the Fuzzy-BPMN miner proposed in this paper) could be considered as a general solution for the pattern discovery International Information Management Association, Inc ISSN: On-line Copy

5 Journal of International Technology and Information Management Volume 26, Number because they obtain better results compared to a single classifier as long as the components are independent and/or have diverse outputs. In turn, this paper trails to make use of such valuable, process-related analysis and capabilities of the PM technique and the classification method to analyse data about a real time business process provided by the IEEE CIS Task Force on Process Mining (Carmona, et al., 2016) in order to show the usefulness and impact of the proposed approach in this paper, namely: the Fuzzy-BPMN miner. In other words, this paper looks at the practical use of such techniques related to the process mining to propose a method that is used to extract meaningful patterns from the event logs captured about those processes, and ways of transforming and analysing the datasets into effective minable formats in order to provide meaningful and worthwhile understanding of the processes as performed in reality. To this end, the work outlines in the following sub-section - the research context and scope of study including the problems which the paper pursues to address and how it is related in context of the research experimentations and proposals. Research questions Primarily, the work in this article explores the best possible ways towards the: RQ1: Use of process mining techniques to discover, monitor and analyse event logs about any domain process by discovering useful and worthwhile process models? RQ2: By what method to determine the extent to which the classification process is able to accurately classify the individual traces that can be found within the event logs and the derived process models? Fundamentally, to address the RQ1 and RQ2, the work utilizes the data about a realtime business process provided by the IEEE CIS Task Force on Process Mining (Carmona, et al., 2016) to show how one can efficiently mine and analyse the sets of unobserved behaviours or patterns (i.e the process instances) that can be found within the event logs in order to discover useful and worthwhile process models. Also, the paper discusses the replaying semantics of the process modelling notations that has been employed, and then provide a description of the tools used to discover the process models as well as evaluation of the results of the classification task. Above all, the work looks at the sophistication of the proposed Fuzzy-BPMN mining approach, validation of the classification tasks, and the discovered process models. International Information Management Association, Inc ISSN: On-line Copy

6 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Research aim The overall goal of the work carried out in this paper is to: extract streams of event logs from any given domain process (case study of the Business Process data from the IEEE CIS Task Force on Process Mining) and describe formats that allows for mining and improved process analysis of the captured data. In other words, the focus of this article is to: apply process mining techniques to a given domain process e.g. the Business Process, and to provide minable formats and understanding about the available datasets (i.e event logs) as well as useful strategies towards the development of process mining techniques/algorithms that exhibits a high level of accuracy for the classification of the individual traces that can be found within the events log and the discovered models. Research objectives Practically, this work uses the case study of the real-time Business Process provided by the IEEE CIS Task Force on Process Mining (Carmona et al, 2016) to seek ways on how to do the following: RO1 Extract data from process domains to show how we harmonize and provide events log formats for any given process domain. RO2 Transform the extracted data into minable executable formats to support the discovery of valuable process models through the proposed technique in this paper. RO3 Provide techniques for accurate classification of unseen process instances (traces) that can be found within the events log and the derived process models. RO4 Assess and evaluate the level of accuracy of the classification process by the proposed method in this paper through further analysis of the discovered models. RO5 Importance of the process mining technique to interpret/support processrelated analysis and enhance information value of data about any domain process: case study of the real-time business process data from the IEEE CIS Task Force on Process Mining. In principle, this article explores the technological potentials and prospects of using the proposed Fuzzy-BPMN mining approach to addresses a typical process International Information Management Association, Inc ISSN: On-line Copy

7 Journal of International Technology and Information Management Volume 26, Number discovery problem in (Carmona, et al., 2016) (as explained in details in the use case scenario and problem statement section of this paper) - by providing a method that combines the capability of the Fuzzy mining algorithms which directly addresses the problem of large numbers of activities and highly unstructured data to show understandable models for very unstructured processes (thus produces simplified process models) and the Business Process Modeling Notation (BPMN) which have proven to be useful towards construction of process models with notational elements that are capable of describing the nesting of individual activities (i.e process instances) by using the event-based split and join gateways (i.e. AND, XOR, and OR etc). Thus, the proposal of the Fuzzy-BPMN miner. In other words, the work introduces by means of the proposed Fuzzy-BPMN miner - a process discovery approach that proves useful towards discovering of new and meaningful process models based on the captured events logs (using the case study of the data about a business process provided by the IEEE CIS Task Force on Process Mining) without any prior information on how those activities are performed. Indeed, the outcome of the research experimentations and data validation (as described in the subsequent experimental section of this paper) shows that the proposed process mining approach has correctly classified to a high percentage the accuracy of the individual traces that can be found within the original process models. Thus, determines the traces which can be replayed (i.e allowed or fitting the model), and the traces which are non-replayable (disallowed or not fitting the model). Accordingly, this article presents the rest of the paper and its results in the following order: 1: Background Information. 2: The Use Case Scenario and Problem Statement. 3: Fuzzy-BPMN Mining Approach: Method, Algorithms & the Classification task. 4: Classified Traces Replay and Model Fitness Calculation 5: Results and Outcome of the Fuzzy-BPMN Mining Approach 6: Discussions and Limitations 7: Conclusion BACKGROUND INFORMATION Process mining (PM) research started at the Eindhoven University of Technology (TU/e) in 1999, and was first proposed by Wil van der Aalst (Van der Aalst, et al., 2003; Van der Aalst, et al., 2004). According to (Van der Aalst, 2016) as of then, there were limited availability of event logs, and the early methods used to perform International Information Management Association, Inc ISSN: On-line Copy

8 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al process mining tasks at that time were exceptionally ineffective and naive. Interestingly, for the past few decades, the process mining tools and approaches has undisputedly matured because event data logs has become ever more available, thanks to the Big Data initiative (Van der Aalst, 2016). Moreover, progress has been spectacular within the process mining field and the technique is currently being supported by various tools and algorithms such as the one introduced in this study. In recent time, the author in (Van der Aalst, 2016) describes the process mining term as one of the main mechanisms of Data Science. The author opines that process mining has the capacity to provide methods towards bridging the gap between data science and process science. According to the author, Process Science has emerged due to the process-perspective that is missing in most big data initiative and the wider curricula of data science. Besides, the author in (Van der Aalst, 2011; Van der Aalst, 2016) argue that the events data logs extracted and stored in many organisations IT system must be utilised to enhance the end-to-end processes in reality by focusing on analysing the unseen behaviours based on the information that are present in the logs, thus, the emergence of process mining. Furthermore, whilst the initial attention was primarily on the process discovery, the PM field have significantly widened, for instance, the conformance checking, operational support, and multi-perspective process mining which has now grown into fundamental part of many tools and approaches that supports the extraction, modelling and/or interpretation of processes. Particularly, ProM (Verbeek, et al., 2011) one of the leading process mining tool currently in literature. Nowadays, several organisations have focused on applying the process mining to different aspects of their business processes management and operations. Moreover, the application of the PM techniques are not only or limited to business processes, but also provides new means to discover, monitor, and enhance any given process domain or interest (De Leoni & Van der Aalst, 2013; Van der Aalst, et al., 2012). According to (Van der Aalst, 2011) there are two main drivers for the growing interest in process mining. First, data about many organizations business processes are captured and stored at an unprecedented rate. Secondly, there is ever increasing need to improve and support business processes in a competitive and rapidly changing environments. This means that - process mining have likewise proved its relatability and application in some other field areas including: Health care (Rojas, et al., 2016), Government sectors (Van der Aalst, 2016), Banking and Financial industries (Jans, 2011; Van der Aalst, et al., 2010), Educational organizations and settings (Cairns, et al., 2014; Okoye, et al., 2016), Airlines and Transportation industry (Van der Aalst, 2016), ICT and Cloud Computing (Chesani, et al., 2016) etc. Indeed, the PM techniques uses event data from any these process domains to discover process models, perform conformance checking International Information Management Association, Inc ISSN: On-line Copy

9 Journal of International Technology and Information Management Volume 26, Number of the discovered models, analyse deviations, and even more, extend and predict future outcomes and/or developments. Actually, many explanations of the PM notion has been propose in literature. Reference (Van der Aalst, 2011) refers to the process mining - as a young research field that makes use of the data mining (DM) technique to find out patterns or models from event logs, and predict outcomes through further analysis of the discovered models. According to the author (Van der Aalst, 2011; Van der Aalst, 2016) PM means extracting valuable, process-related information from event logs about any domain process. Reference (Cairns, et al., 2015) also mentions that the PM term is concerned with analysis of captured events log from a process-perspective. Reference (Ingvaldsen, 2011) states that as soon as a particular process (e.g. business process) is being supported by some form of IT system, its operational transactions or activities executions can then be observed or recorded in the form of event logs. Likewise, references (Greco, et al., 2006; Van der Aalst, 2011) notes that the process mining notion is an attempt towards extraction of meaningful and non-trivial information from recorded event logs. Notably, the lion s share of attention in process mining has been devoted to the process discovery - i.e., extracting process models, mainly business process models from recorded events log (Carmona, et al., 2016). The Process discovery has been lately seen as the main significant and furthermost challenge logically allied to the PM term (Carmona, et al., 2016; Van der Aalst, 2011). Process discovery techniques aims to automatically construct process models, e.g., BPMN, Petri-nets, C-nets, Fuzzy models, Process Trees etc. (Van der Aalst, 2016) from events log about a process, and describes causal dependencies between the individual activities as performed in reality. In short, a typical process discovery method takes (as input) recorded event logs, and then produce (as output) a model without any prior information on how the activities has been formerly performed. Besides, in settings where the datasets (i.e. event logs) includes information about resource (e.g. roles), it is also possible to discover resource-related models. For instance, a shared neural network model representing how employees works collectively or collaborate within a particular organisation. In essence, one can make use of the process discovery methods to obtain models that describes reality. More so, the conformance checking is the second type of process mining techniques. The method focuses on determining (assessing) how fit the discovered process models describes the actual observation in the event logs (Ingvaldsen, 2011) such as the approach described in this paper. Principally, a conformance International Information Management Association, Inc ISSN: On-line Copy

10 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al check and analysis technique references an a-priori (i.e. existing) process model and compares it with the events log of the specific (i.e. the same) process. Clearly, such analysis is performed in order to check if in reality, the recorded events data log conforms to the deployed process models (Munoz-Gama & Carmona, 2011; Adriansyah, et al., 2011; Rozinat & Van der Aalst, 2008; Weerdt, et al., 2011; Fahland & van der Aalst, 2012). For instance, the output a conformance checking technique may imply that the discovered models perhaps do not describe the executed process as supposed in reality, or is being executed in a different order (Fahland & van der Aalst, 2012; Van der Aalst, 2011). It could also mean that some of the process instance (i.e. individual activities) as observed within the discovered model are skipped in the events log, or may be the logs consist of actions (i.e. events) that are not necessarily defined by the process model (Fahland & van der Aalst, 2012). Therefore, a well performed conformance check is relevant and significant especially from a business objective alignment or auditing perspective. For example, it is possible that the recorded logs could be reiterated (i.e. model replay) against the derived models in order to discover unexpected deviation or bottlenecks that may impact the entire business process in general. In other words, the conformance checking could be utilized to measure the fitness of the models discovered by the PM tools. For instance, the level or extent of behaviours within the event logs which happen to be actually possible according to the discovered process models, and could also be used to perform the repairing of the models in reality. In fact, the conformance checking technique is utilised to balance between traces (i.e. observed behaviours or patterns within the events logs and models) that are overfitting or underfitting the actual process as performed in reality (Carmona, et al., 2016; Fahland & van der Aalst, 2012). According to (Van der Aalst, 2016), most often conformance check is performed to show the replaying semantics (or better still - token replay) for models with regards to the four quality criteria s - Fitness, Generalisation, Precision, and Simplicity (Van der Aalst, 2011). In summary, the process mining plays an important role in many organisations. It spans its technical application from the fields of data science and business process management (BPM), and as such, we assume that to perform any process mining task that there has to be some kind of recorded data from an actual process. For instance, as this study uses data from the IEEE CIS Task Force on Process Mining (Carmona, et al., 2016) to perform the process models discovery and individual traces classification: which are explained in details in the subsequent sections of this paper. Also, using the Learning Process domain for example, the Figure 1 shows that the first step (i.e. starting point) for any given process mining project is to capture the event data logs about the process, and then generate process model International Information Management Association, Inc ISSN: On-line Copy

11 Journal of International Technology and Information Management Volume 26, Number to show in details how the activities has been performed and to reveal interesting connections between the different process elements (i.e the process instances). In turn, the process mappings can subsequently be utilized to provide methods that allows for an enhanced analysis and/or extension of the discovered process model. Thus, the three types of the process mining techniques Process Discovery, Conformance Check, and Model Enhancement. Figure 1. Application of the process mining techniques USE CASE SCENARIO AND PROBLEM STATEMENT The proposed Fuzzy-BPMN miner approach in this paper is directed towards the discovery of process models from a set of Training Event Logs representing 10 different real-time Business Process executions, and then followed by crossvalidation of the derived models with a set of Test Event Logs used for evaluation of the process discovery technique and the accuracy of the classification method. Each of the test event logs precisely ((test_log_april_1 to test_log_april_10) and (test_log_may_1 to test_log_may_10)) which can be found in (Carmona, et al., 2016) represents part of the original model as recorded by the IEEE CIS Task Force on Process Mining. Also, the test logs with complete total of 20 traces for each log are considered to consist of 10 traces which are replayable (i.e. allowed) and another 10 traces which are not replayable (disallowed) by the model. Therefore, the total number of traces for the test event logs (i.e. April log and May log) is thus: International Information Management Association, Inc ISSN: On-line Copy

12 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al 10 test logs x 20 traces which equals to a total number of = 200 Traces for each of the April log and May log respectively Clearly, the aim of the work carried out in this paper is to perform a classification task in order to determine the individual traces that makes up the test event logs, and then provide process models with high fitness levels using the Business Process Model Notation (BPMN) mapping for the training event logs which allows for testing and evaluation of the classified traces (i.e. the discovered patterns) observed in the test logs. In other words, the objective of the proposed approach is to discover and provide process models that matches the original process models in term of balancing between overfitting and underfitting. On the one hand, a process model is seen as overfitting (the event log) if it is too restrictive, disallowing behaviour which is part of the underlying process. On the other hand, it is underfitting (the reality) if it is not restrictive enough, allowing behaviour which is not part of the underlying process. Therefore, following the problem statement and objectives, this article focuses on providing process models which is as good in balancing between overfitting and underfitting as it is able to correctly classify the traces that can be replayed on the model or not replayable based on the analysis of the classification results. Thus, Given a trace (t) representing real process behaviour, the process model (m) classifies it as allowed, or Given a trace (t) representing a behaviour not related to the process, the process model (m) classifies it as disallowed (Carmona, et al., 2016) In summary, the work in this paper covers the classification attempts for the events logs provided in (Carmona, et al., 2016) and discusses the replaying semantics of the process modelling notation that has been employed. Hence, we reveal how given any process trace (t) (for the test event log) and process model (m) (for the training log) in the employed Fuzzy mining and BPMN notation, it can be unambiguously determined whether or not trace (t) can be replayed on model (m). The study also provides a description of the tools used to discover the process models as well as in checking the result of the classification task. In fact, the method the work has utilized to resolve the identified problem and challenge is grounded on the proposed Fuzzy-BPMN mining approach and Algorithm as described in the following section. International Information Management Association, Inc ISSN: On-line Copy

13 Journal of International Technology and Information Management Volume 26, Number FUZZY-BPMN MINING APPROACH: METHOD, ALGORITHMS & THE CLASSIFICATION TASK This section of the paper describes the proposed algorithm and method the work have used to perform the classification task of the Event Data Logs using the Fuzzy- BPMN approach. The method is applied in order to generate the individual traces that makes up each of the process executions as described in (Carmona et al., 2016). In addition, we show how we implement the proposed approach using PM tools such as the Disco (Rozinat & Gunther, 2012) and ProM (Verbeek, et al., 2011; Verbeek, 2014). It is important to note that the proposed algorithm and the defined method of this paper is independent on which tool one may choose in order to analyse the available datasets. Moreover, the work has carefully assessed both by hand and in an automatic manner the performance of the proposed system particularly for comparison and validation purposes. For instance, the work has used the Disco tool based on the Fuzzy Miner algorithm to generate and map the process models (from the event logs) for conformance checking and further analysis: which allows us to automatically determine the individual Cases i.e. the classified traces (20 for each Log) and the sequence of activity executions as performed within the process in reality. And, on the other hand, has carefully cross-validated the results of the classification task (see: Table 1) against the resultant BPMN models that were derived from the training logs. Indeed, the procedures are all aimed at ascertaining the level of performance and accuracy of the classification results of the proposed Fuzzy-BPMN mining approach, particularly in terms of the individual traces and the discovered models. In turn, the following Algorithm 1 describes how this work discovers and generates (i) process models and (ii) individual traces - from any event data log containing a Training sets and Test Logs respectively. Algorithm 1: Discovering Process Models from Event Logs & Traces Classification 1: For all Recorded and Captured Event Data Log EDL 2: Input: PM Process mining tool used to extract model, M e Classifier for the event logs, EDL and traces, T case_id(e) - i.e. the Case associated to any event, e within the EDL act_name(e)- i.e. the Activity associated to event, e within the EDL 3: Output: Process maps for the discovered models, M & individuals traces, T classifications for the event log, EDL International Information Management Association, Inc ISSN: On-line Copy

14 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al 4: Procedure: Produce Models, M from Training Set, TSL and Traces, T from Test Log, TEL for cross-validation to determine the model tracefitness, TF 5: Begin 6: For all Event Data Log EDL 7: Extract Process Maps, M, & Traces, T from Training Set, TSL & Test Log, TEL 8: while no more process element is left do 9: Analyze Model, M and Traces, T to determine individual tracesfitness, TF 10: If T Null then 11: obtain the occurring activities act_name(e) sequence sets from test log, TEL 12: Else If T 1 then 13: cross-validate resulting Trace, T from TEL with discovered Model, M from TSL 14: If trace, T exist then 15: For each event Classifier, e output return as True_Positive, TP 16: Else If trace, T does not exist then return event Classifier, e output as True_Negative, TN 17: Record the tracefitness, TF in Table as True or False: where each individual cell indicates if the discovered model classifies the corresponding trace as fitting (allowed i.e. TP) or not fitting (disallowed i.e. TN). 18: Return: Classification Results of the Experiment and Process Mining approach 19: End If statements 20: End while 21: End For Ultimately, from the proposed Algorithm 1, we recognises that: A typical process model, M consist of Traces, T (i.e. Cases) A Trace (Case), T, consist of events, e, such that each event relates to precisely one case. Events, e, within a Trace are ordered, most often in a sequential order Events for any process mining task must have atleast a Case identification ID (i.e Case_id) and Activity Name (i.e Act_name) attributes to allow for the process model discovery. Other additional information may be required for ample implementation of the process mining technique e.g. Event ID, Timestamp, Resources, Cost, Roles, Places etc. International Information Management Association, Inc ISSN: On-line Copy

15 Journal of International Technology and Information Management Volume 26, Number Accordingly, the event log that have been provided by the IEEE CIS Task Force on Process Mining (Carmona, et al., 2016) for the models discovery process contains the typical information needed for process mining particularly in achieving the focus of this article in terms of the process models discovery and implementation of the proposed Algorithm 1. The provided Datasets represents and shows events logs generated from a business process model to show different behavioural characteristics. We assume that each of the events log contains data related to at least a single process which also refers to a single process instance (i.e. Case) and can be related to some Activity. Moreover, according to (Van der Aalst, 2011) a Case ID and Activity is the minimum requirement for any process mining technique. Indeed, the given event logs in (Carmona, et al., 2016) contains the two attributes - case_id and act_name which precisely specify the requirements that allows for implementation of the proposed process discovery technique in this paper, especially in line with the definition 4.1 in (Van der Aalst, 2011). Therefore, we assume the following standard: #case_id(e) is the Case associated to any event e. #act_name(e) is the Activity associated to event e. These definitions are necessary because for the Fuzzy-BPMN miner approach - the activities play an important role in terms of the discovered models, and thus, is used to check for the corresponding cases (i.e. classified traces) within the models. Even more, as there are multiple events referring to the same Activity, we support the filtering of the 200 individual traces (each for the April and May logs) that makes up the test event logs with a classifier as described in definition 4.2 in (Van der Aalst, 2011). According to (Van der Aalst, 2011) a classifier is a function that maps the attributes of an event onto a label used in the resulting process model. Thus, if we use the notation e to refer to the events name within the process model, then the classifier for any event in the given Log will be, e E, where e is the name of the event. More so, since the events are simply identified by their activity name (act_name), we then assume: e = #act_name(e) Finally, we apply the classification conversion of the event logs provided (i.e. Simple Event Log, see: Definition 4.4) in (Van der Aalst, 2011) to obtain the individual Log traces. International Information Management Association, Inc ISSN: On-line Copy

16 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Therefore, applying the described simple event log definition: Let A be a set of act_name. A simple/single trace σ is a sequence of activities, i.e., σ A *. In other words, a simple event Log, L, is a multiset of traces over some set A. Thus, L B (A* ). On the other hand, for the Training Log there are 1000 cases (trace) that defines the log. However, our focus is to identify the sets of traces (i.e. 200 for April and 200 for May logs respectively) that characterize the Test Log for use in validation of the process model discovery method in this paper, particularly the April Logs which were used to score the number of correctly classified traces as well as the experimental outcomes. Therefore, If we Let L C be the event log for the Test Logs, and assuming that the classifier e E, is applied to the set of sequences of the activities, then from the definition (4.5) in (Van der Aalst, 2011) e1, e2,, en = e1, e2,, en where L = [(ĉ) c L ] is a simple event log corresponding to Test Log. All the Cases in the Test Log are converted into sequences of the activities (act_name) using the classifier. Hence A Case c L, is an identifier from the case C. ĉ = #trace(c) = e1, e2,, en E is the sequence of events executed for c (ĉ) = e1, e2,, en maps these events onto the activity names(act_name) using the classifier. Thus, from the described classification method: (e = #act_name(e)), we obtain from the Log containing the set of 200 traces for the Test Event Log (test_log_april_1) to (test_log_april_10), i.e., 20 Traces for each log as follows: L (test_log_april_1) = [ b, g, e, q, h, i, l, r, m, o, d, f, p, b, b, c, n, h, e, i, q, r, l, m, f, o, d, p, g, h, I, q, q, m, r, o, e, d, p, j, a, k, b, b, g, e, h, q, l, r, i, m, d, f, o, p, b, g, h, i, q, i, r, m, o, d, p, f, e, e, e, q, h, r, d, o, r, p, g, h, e, i, i, q, l, m, o, f, p, d, International Information Management Association, Inc ISSN: On-line Copy

17 Journal of International Technology and Information Management Volume 26, Number b, a, j, k, g, e, q, h, l, i, r, m, o, f, d, p, g, i, e, r, l, i, m, d, o, p, d, p, b, b, g, e, l, l, h, q, r, r, r, d, o, o, p, f, b, g, e, h, i, q, l, r, m, d, p, o, f, b, q, g, h, i, h, l, m, m, r, p, f, h, g, h, e, r, l, q, i, f, f, p, b, j, a, k, g, q, e, i, h, l, r, f, d, o, p, c, n, q, e, i, h, r, d, m, o, p, f, p, b, g, h, i, e, q, r, l, m, d, o, p, f, g, i, h, e, r, q, m, l, o, d, f, p, k, b, n, n, c, h, h, e, q, l, q, r, r, i, m, f, f, i, p, b, b, b, g, q, i, h, e, r, l, m, f, o, d, p, b, b, g, q, e, h, i, r, m, l, d, o, p, f ] The Log; L (test_log_april_1) is example of the set of 20 traces which the work obtained for the test_log_april_1. Further details of all the classified traces for the complete test logs can be found in (Okoye, et al., 2016). EXPERIMENTAL SET-UP AND CONFORMANCE CHECKING OF THE INDIVIDUAL CLASSIFIED TRACES The Event Logs used for the process models discovery in this article has been provided in XES (Extensible Event Streams) format. A typical XES document contains logs which consist of traces. Each trace describes a sequential list of events corresponding to a particular case in terms of the concept:name, for instance, the case_id and act_name attributes. The XES files refers to these extensions to provide semantics for the Logs. Truly, in recent times the most widely standard for storing and exchanging event logs across different platforms for process mining is the XES; because the format is less restrictive and truly extendible. Interestingly, XES has been adopted by the IEEE CIS Task Force on Process Mining since 2010 as standard format for process mining and is supported by tools such as the ProM (Verbeek, et al., 2011; Verbeek, 2014), Disco (Rozinat & International Information Management Association, Inc ISSN: On-line Copy

18 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Gunther, 2012), XE-Same (Verbeek, et al., 2011), OpenXES (Gunther & Verbeek, 2014) etc. Furthermore, a typical attribute for the XES format can be of five core types namely: String, Date, Int, Float, and Boolean. For instance, the case_id, act_name which are of StringType. Moreover, these extensions gives semantics for a particular attribute. For example, the extensions corresponds to the #case_id(e) and #act_name(e) attribute which we used to classify the traces for the test logs. In fact, there are three classifiers defined by XES which are as follows: (i) (ii) (iii) Classifier Activity (concept:name), Classifier Resource (org:resource), Classifier Both (concept:name and org:resource). Nonetheless, for the purpose of the method and experimentations carried in this paper: our focus is on the Classifier Activity because the main objective is to classify the events in the test logs based on the concept:name attributes i.e. act_name, and case_id for each of the Event Name in terms of their string values and order/sequence of the Lifecycle transition (as shown in the highlighted part of the Figure 2) and then cross validate the resulting traces with the training set (i.e. the discovered models). Indeed, XES supports the classifier concept and as such helps in specifying the list of the attributes associated with the concept:name as gathered in Figure 2. International Information Management Association, Inc ISSN: On-line Copy

19 Journal of International Technology and Information Management Volume 26, Number Figure 2. Fragment of the XES file format for the test event log International Information Management Association, Inc ISSN: On-line Copy

20 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Following the definitions in the above section and Figure 2, we classified the test event logs. More so, we imported the XES files for the Test Logs into Disco (Rozinat & Gunther, 2012) as shown in Figure 3 to see in details how those processes has been performed (i.e. the Process mapping), and more importantly to determine the individual Cases (trace) that makes up the process in order to check if it matches with the classified traces. Figure 3. Event Log analysis using the Fuzzy miner algorithm in Disco. In Figure 3 we assigned the ID Tag to the first column (i.e case_id) in order to identify the events, and the Activity Tag to the second column (i.e act_name) to determine the sets of activity that makes up the process. Apparently, the outcome of the process is a fuzzy model that represents the various cases and activities sequence mapping for the events log as shown in Figure 4 and 5. Figure 4. Case View for the test_log_april_1 showing the 20 cases and graph for the activities sequence. International Information Management Association, Inc ISSN: On-line Copy

21 Journal of International Technology and Information Management Volume 26, Number Figure 5. Case view for the test_log_april_1 showing the 20 cases with an example of case 1 (trace) with 13 events and sets of Activity for trace 1. Indeed, the approach described in Figure 4 and 5 is what we used to check the results of the classification tasks; to see if the outcome of the process confirms to the given event logs. For example, the activities for the first case 1 as highlighted in Figure 5, truly corresponds to the first trace discovered by the classifier, i.e. L (test_log_april_1) = [ b, g, e, q, h, i, l, r, m, o, d, f, p, etc. To this end, and in view of the individual traces classification results, we make use of the proposed Fuzzy-BPMN mining approach to determine the fitness (replaying semantics) of the individual traces for the test event logs by cross-validating the classified traces against the discovered process models from the training logs as discussed in the next section of this paper. THE PROCESS MODELS DISCOVERY METHOD AND ANALYSIS To discover process models for the event logs (i.e the training logs) used for the experimentations, the work makes use of the Fuzzy miner algorithm in Disco (Rozinat & Gunther, 2012) to process the data. At first, the work discovers 10 different process models from the training sets (Carmona, et al., 2016) using the Fuzzy miner (G unther, 2009; Günther & Van der Aalst, 2007; Rozinat & Gunther, 2012) and then subsequently utilize the Business Process Modeling Notation (BPMN) (Van der Aalst, 2016) to analyze and provide the replaying semantics for International Information Management Association, Inc ISSN: On-line Copy

22 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al the process models. Figure 6 is an example of the discovered models for the traning_log_1 that was analyzed. Further details about the 10 different process models that were discovered and analyzed using the proposed method is described in (Okoye, et al., 2016) and are provided in the Appendix A section of this article. Figure 6. Fuzzy Model discovered for the Training_Log_1 CLASSIFIED TRACES REPLAY AND MODEL FITNESS CALCULATION Process Mining aims to address the problem of establishing a direct connection between discovered models and the actual low-level event data about the processes in view. Besides, the process discovery techniques allows for viewing the same reality from different angles and at different levels of abstraction. To evaluate and cross-validate the classification tasks for the test event log (i.e April Log) with the training model, we base our technique towards balancing between overfitting and underfitting models as described in section in (Van der Aalst, 2011) - which focuses on expending measures of data performance indicator using the four quality criterias: Fitness, Precision, Generalisation and Simplicity as shown in Figure 7. International Information Management Association, Inc ISSN: On-line Copy

23 Journal of International Technology and Information Management Volume 26, Number Figure 7. Four competing quality criteria for evaluation of process models (Van der Aalst, 2011) As gathered in Figure 7, we consider the four quality criteria to explain the level of accuracy (particularly fitness) of the discovered models as defined in section 3.6 in (Van der Aalst, 2011) in order to determine which fractions of the traces in the test logs can be fully replayed or are disallowed by the discovered models. Thus: Fitness: the discovered model should allow for behaviours seen in the event log. Thus, is the event log possible according to the discovered model? Precision (avoiding underfitting): the discovered model should not allow for behaviours completely unrelated to what was seen in the event logs. Thus, is the model not underfitting i.e. allows for too much? Generalization (avoiding overfitting): the discovered model should generalize the example behaviours seen in the event logs. Thus, is the model not overfitting i.e. only allows for particular examples? Simplicity (Occam s razor principle): the discovered model should be as simple as possible. Thus, is the discovered model the simplest? One should not increase, beyond what is necessary, the number of entities required to explain anything, i.e., one should look for the simplest model that can explain what is observed in the dataset. Essentially, the fitness of the discovered models is judged on the Training Logs which are measured against the test logs classification results as shown in Figure 8 - also referred to as Cross-Validation in section in (Van der Aalst, 2011). International Information Management Association, Inc ISSN: On-line Copy

24 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Figure 8. Cross-validation using a training set and test set (Van der Aalst, 2011) Furthermore, according to (Van der Aalst, 2011), the conformance checking is closely related to measuring the fitness of the discovered models, and it can also be used to evaluate and compare the process discovery algorithms. Section 7.2 of (Van der Aalst, 2011) discusses the replaying semantics (Token Replay) for the process models with respect to the four quality criteria. The token replay shows how the notion of event log fitness can be quantified i.e. the proportion of behaviours in the event logs that are possible according to the discovered models. In other words, the token replay are used to establish a tight coupling between the discovered model and the event logs. For that reason, to achieve the set objective of the paper - it was necessary to construct BPMN models with notational elements (as explained in Figure 9) capable of describing the nesting of individual activities (traces) by using the eventbased split and join gateways, i.e. AND, XOR, and OR etc. Moreover, since our target is to classify as correctly as possible the traces which are allowed and the traces which are not allowed in the original model, the work utilized the BPMN event-based gateways to replay the individual traces fitness alongside the derived models from the training event logs, and in so doing, identify which traces that are fitting or not fitting the original model. International Information Management Association, Inc ISSN: On-line Copy

25 Journal of International Technology and Information Management Volume 26, Number Figure 9. BPMN Gateway with Notational elements (Van der Aalst, 2011) Indeed, an event in a BPMN model can be compared to a place within a Petri-net (Van der Aalst, 2011), and just like Petri nets, are token based semantics which can be used to replay a particular trace within a discovered process model (Van der Aalst, 2016). To this end, the work makes use of the Convert Petri net to BPMN plugin in ProM (Verbeek, et al., 2011) to discover the BPMN models for the training logs. Figure 10 is an example of the discovered BPMN Diagram for the training_log_1. Further details about the other 10 different BPMN models that was discovered using the method can be found in (Okoye, et al., 2016) and also included in the Appendix B section of this paper. Figure 10. Example of BPMN model discovered for the training_log_1 Consequently, in Table 1 the study presents the classification results of the Fuzzy- BPMN mining approach for the test event logs cross-validated against the corresponding training set (model): where each individual cell indicates if the discovered model classifies the corresponding trace as fitting (i.e allowed) or not fitting (disallowed). Thus, the columns represents the process models for the 10 training logs, while the rows represents the individual traces for the test log. For example, cell at row Trace_3 column Training model_5 contains the classification attempt for the 3rd trace discovered from the test_log_april_5 crossvalidated against the training_log_5. International Information Management Association, Inc ISSN: On-line Copy

26 Process Models Discovery and Traces Classification: A Fuzzy-BPMN Mining Approach Kingsley Okoye et al Table 1. Classified Trace fitness Table for the test event logs (test_log_april_1 to test_log_april_10) RESULTS AND OUTCOME OF THE FUZZY-BPMN MINING APPROACH The IEEE CIS Task Force on Process Mining contest committee published on the website (Carmona, et al., 2016) - (a) 10 test logs, each of which contains 20 traces that were used to score the submission report, and (b) 10 reference process models in BPMN generated from the original event logs which were not previously revealed. The Table 2 represents the final results and scoring of the employed Fuzzy-BPMN mining approach and experimentations in this paper. International Information Management Association, Inc ISSN: On-line Copy

27 Journal of International Technology and Information Management Volume 26, Number Table 2. Trace Fitness and Classification Table for the Test Event Logs (test_log_april_1 to test_log_april_10) using the Fuzzy-BPMN Miner Consequently, as shown in Table 2 the following performance metrics (Van der Aalst, 2011; Van der Aalst, 2016) were used to measure the fitness of the individual traces for the datasets, where: TP is the number of true positives i.e. instances that are correctly classified as positive FN is the number of false negatives i.e. instances that are predicted to be negative but should have been classified as positive FP is the number of false positives i.e. instances that are predicted to be positive but should have been classified as negative TN is the number of true negatives (i.e. instances that are correctly classified as negative) Accordingly, the cells with gold sign (*) indicates the traces that were correctly classified by the Fuzzy-BPMN miner after scoring the classification results and models. Indeed, the final result after scoring by the committee experts in process International Information Management Association, Inc ISSN: On-line Copy